kernel/fs/proc/task_mmu.c

   1 #include <linux/mm.h>
   2 #include <linux/vmacache.h>
   3 #include <linux/hugetlb.h>
   4 #include <linux/huge_mm.h>
   5 #include <linux/mount.h>
   6 #include <linux/seq_file.h>
   7 #include <linux/highmem.h>
   8 #include <linux/ptrace.h>
   9 #include <linux/slab.h>
  10 #include <linux/pagemap.h>
  11 #include <linux/mempolicy.h>
  12 #include <linux/rmap.h>
  13 #include <linux/swap.h>
  14 #include <linux/swapops.h>
  15 #include <linux/mmu_notifier.h>
  16 #include <linux/page_idle.h>
  17
  18 #include <asm/elf.h>
  19 #include <asm/uaccess.h>
  20 #include <asm/tlbflush.h>
  21 #include "internal.h"
  22
  23 void task_mem(struct seq_file *m, struct mm_struct *mm)
  24 {
  25         unsigned long data, text, lib, swap, ptes, pmds;
  26         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  27
  28         /*
  29          * Note: to minimize their overhead, mm maintains hiwater_vm and
  30          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  31          * collector of these hiwater stats must therefore get total_vm
  32          * and rss too, which will usually be the higher.  Barriers? not
  33          * worth the effort, such snapshots can always be inconsistent.
  34          */
  35         hiwater_vm = total_vm = mm->total_vm;
  36         if (hiwater_vm < mm->hiwater_vm)
  37                 hiwater_vm = mm->hiwater_vm;
  38         hiwater_rss = total_rss = get_mm_rss(mm);
  39         if (hiwater_rss < mm->hiwater_rss)
  40                 hiwater_rss = mm->hiwater_rss;
  41
  42         data = mm->total_vm - mm->shared_vm - mm->stack_vm;
  43         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
  44         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
  45         swap = get_mm_counter(mm, MM_SWAPENTS);
  46         ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
  47         pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
  48         seq_printf(m,
  49                 "VmPeak:\t%8lu kB\n"
  50                 "VmSize:\t%8lu kB\n"
  51                 "VmLck:\t%8lu kB\n"
  52                 "VmPin:\t%8lu kB\n"
  53                 "VmHWM:\t%8lu kB\n"
  54                 "VmRSS:\t%8lu kB\n"
  55                 "VmData:\t%8lu kB\n"
  56                 "VmStk:\t%8lu kB\n"
  57                 "VmExe:\t%8lu kB\n"
  58                 "VmLib:\t%8lu kB\n"
  59                 "VmPTE:\t%8lu kB\n"
  60                 "VmPMD:\t%8lu kB\n"
  61                 "VmSwap:\t%8lu kB\n",
  62                 hiwater_vm << (PAGE_SHIFT-10),
  63                 total_vm << (PAGE_SHIFT-10),
  64                 mm->locked_vm << (PAGE_SHIFT-10),
  65                 mm->pinned_vm << (PAGE_SHIFT-10),
  66                 hiwater_rss << (PAGE_SHIFT-10),
  67                 total_rss << (PAGE_SHIFT-10),
  68                 data << (PAGE_SHIFT-10),
  69                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
  70                 ptes >> 10,
  71                 pmds >> 10,
  72                 swap << (PAGE_SHIFT-10));
  73         hugetlb_report_usage(m, mm);
  74 }
  75
  76 unsigned long task_vsize(struct mm_struct *mm)
  77 {
  78         return PAGE_SIZE * mm->total_vm;
  79 }
  80
  81 unsigned long task_statm(struct mm_struct *mm,
  82                          unsigned long *shared, unsigned long *text,
  83                          unsigned long *data, unsigned long *resident)
  84 {
  85         *shared = get_mm_counter(mm, MM_FILEPAGES);
  86         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  87                                                                 >> PAGE_SHIFT;
  88         *data = mm->total_vm - mm->shared_vm;
  89         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
  90         return mm->total_vm;
  91 }
  92
  93 #ifdef CONFIG_NUMA
  94 /*
  95  * Save get_task_policy() for show_numa_map().
  96  */
  97 static void hold_task_mempolicy(struct proc_maps_private *priv)
  98 {
  99         struct task_struct *task = priv->task;
 100
 101         task_lock(task);
 102         priv->task_mempolicy = get_task_policy(task);
 103         mpol_get(priv->task_mempolicy);
 104         task_unlock(task);
 105 }
 106 static void release_task_mempolicy(struct proc_maps_private *priv)
 107 {
 108         mpol_put(priv->task_mempolicy);
 109 }
 110 #else
 111 static void hold_task_mempolicy(struct proc_maps_private *priv)
 112 {
 113 }
 114 static void release_task_mempolicy(struct proc_maps_private *priv)
 115 {
 116 }
 117 #endif
 118
 119 static void vma_stop(struct proc_maps_private *priv)
 120 {
 121         struct mm_struct *mm = priv->mm;
 122
 123         release_task_mempolicy(priv);
 124         up_read(&mm->mmap_sem);
 125         mmput(mm);
 126 }
 127
 128 static struct vm_area_struct *
 129 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
 130 {
 131         if (vma == priv->tail_vma)
 132                 return NULL;
 133         return vma->vm_next ?: priv->tail_vma;
 134 }
 135
 136 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
 137 {
 138         if (m->count < m->size) /* vma is copied successfully */
 139                 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL;
 140 }
 141
 142 static void *m_start(struct seq_file *m, loff_t *ppos)
 143 {
 144         struct proc_maps_private *priv = m->private;
 145         unsigned long last_addr = m->version;
 146         struct mm_struct *mm;
 147         struct vm_area_struct *vma;
 148         unsigned int pos = *ppos;
 149
 150         /* See m_cache_vma(). Zero at the start or after lseek. */
 151         if (last_addr == -1UL)
 152                 return NULL;
 153
 154         priv->task = get_proc_task(priv->inode);
 155         if (!priv->task)
 156                 return ERR_PTR(-ESRCH);
 157
 158         mm = priv->mm;
 159         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
 160                 return NULL;
 161
 162         down_read(&mm->mmap_sem);
 163         hold_task_mempolicy(priv);
 164         priv->tail_vma = get_gate_vma(mm);
 165
 166         if (last_addr) {
 167                 vma = find_vma(mm, last_addr);
 168                 if (vma && (vma = m_next_vma(priv, vma)))
 169                         return vma;
 170         }
 171
 172         m->version = 0;
 173         if (pos < mm->map_count) {
 174                 for (vma = mm->mmap; pos; pos--) {
 175                         m->version = vma->vm_start;
 176                         vma = vma->vm_next;
 177                 }
 178                 return vma;
 179         }
 180
 181         /* we do not bother to update m->version in this case */
 182         if (pos == mm->map_count && priv->tail_vma)
 183                 return priv->tail_vma;
 184
 185         vma_stop(priv);
 186         return NULL;
 187 }
 188
 189 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 190 {
 191         struct proc_maps_private *priv = m->private;
 192         struct vm_area_struct *next;
 193
 194         (*pos)++;
 195         next = m_next_vma(priv, v);
 196         if (!next)
 197                 vma_stop(priv);
 198         return next;
 199 }
 200
 201 static void m_stop(struct seq_file *m, void *v)
 202 {
 203         struct proc_maps_private *priv = m->private;
 204
 205         if (!IS_ERR_OR_NULL(v))
 206                 vma_stop(priv);
 207         if (priv->task) {
 208                 put_task_struct(priv->task);
 209                 priv->task = NULL;
 210         }
 211 }
 212
 213 static int proc_maps_open(struct inode *inode, struct file *file,
 214                         const struct seq_operations *ops, int psize)
 215 {
 216         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 217
 218         if (!priv)
 219                 return -ENOMEM;
 220
 221         priv->inode = inode;
 222         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 223         if (IS_ERR(priv->mm)) {
 224                 int err = PTR_ERR(priv->mm);
 225
 226                 seq_release_private(inode, file);
 227                 return err;
 228         }
 229
 230         return 0;
 231 }
 232
 233 static int proc_map_release(struct inode *inode, struct file *file)
 234 {
 235         struct seq_file *seq = file->private_data;
 236         struct proc_maps_private *priv = seq->private;
 237
 238         if (priv->mm)
 239                 mmdrop(priv->mm);
 240
 241         return seq_release_private(inode, file);
 242 }
 243
 244 static int do_maps_open(struct inode *inode, struct file *file,
 245                         const struct seq_operations *ops)
 246 {
 247         return proc_maps_open(inode, file, ops,
 248                                 sizeof(struct proc_maps_private));
 249 }
 250
 251 static pid_t pid_of_stack(struct proc_maps_private *priv,
 252                                 struct vm_area_struct *vma, bool is_pid)
 253 {
 254         struct inode *inode = priv->inode;
 255         struct task_struct *task;
 256         pid_t ret = 0;
 257
 258         rcu_read_lock();
 259         task = pid_task(proc_pid(inode), PIDTYPE_PID);
 260         if (task) {
 261                 task = task_of_stack(task, vma, is_pid);
 262                 if (task)
 263                         ret = task_pid_nr_ns(task, inode->i_sb->s_fs_info);
 264         }
 265         rcu_read_unlock();
 266
 267         return ret;
 268 }
 269
 270 static void
 271 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
 272 {
 273         struct mm_struct *mm = vma->vm_mm;
 274         struct file *file = vma->vm_file;
 275         struct proc_maps_private *priv = m->private;
 276         vm_flags_t flags = vma->vm_flags;
 277         unsigned long ino = 0;
 278         unsigned long long pgoff = 0;
 279         unsigned long start, end;
 280         dev_t dev = 0;
 281         const char *name = NULL;
 282
 283         if (file) {
 284                 struct inode *inode = file_inode(vma->vm_file);
 285                 dev = inode->i_sb->s_dev;
 286                 ino = inode->i_ino;
 287                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 288         }
 289
 290         /* We don't show the stack guard page in /proc/maps */
 291         start = vma->vm_start;
 292         if (stack_guard_page_start(vma, start))
 293                 start += PAGE_SIZE;
 294         end = vma->vm_end;
 295         if (stack_guard_page_end(vma, end))
 296                 end -= PAGE_SIZE;
 297
 298         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 299         seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
 300                         start,
 301                         end,
 302                         flags & VM_READ ? 'r' : '-',
 303                         flags & VM_WRITE ? 'w' : '-',
 304                         flags & VM_EXEC ? 'x' : '-',
 305                         flags & VM_MAYSHARE ? 's' : 'p',
 306                         pgoff,
 307                         MAJOR(dev), MINOR(dev), ino);
 308
 309         /*
 310          * Print the dentry name for named mappings, and a
 311          * special [heap] marker for the heap:
 312          */
 313         if (file) {
 314                 seq_pad(m, ' ');
 315                 seq_file_path(m, file, "\n");
 316                 goto done;
 317         }
 318
 319         if (vma->vm_ops && vma->vm_ops->name) {
 320                 name = vma->vm_ops->name(vma);
 321                 if (name)
 322                         goto done;
 323         }
 324
 325         name = arch_vma_name(vma);
 326         if (!name) {
 327                 pid_t tid;
 328
 329                 if (!mm) {
 330                         name = "[vdso]";
 331                         goto done;
 332                 }
 333
 334                 if (vma->vm_start <= mm->brk &&
 335                     vma->vm_end >= mm->start_brk) {
 336                         name = "[heap]";
 337                         goto done;
 338                 }
 339
 340                 tid = pid_of_stack(priv, vma, is_pid);
 341                 if (tid != 0) {
 342                         /*
 343                          * Thread stack in /proc/PID/task/TID/maps or
 344                          * the main process stack.
 345                          */
 346                         if (!is_pid || (vma->vm_start <= mm->start_stack &&
 347                             vma->vm_end >= mm->start_stack)) {
 348                                 name = "[stack]";
 349                         } else {
 350                                 /* Thread stack in /proc/PID/maps */
 351                                 seq_pad(m, ' ');
 352                                 seq_printf(m, "[stack:%d]", tid);
 353                         }
 354                 }
 355         }
 356
 357 done:
 358         if (name) {
 359                 seq_pad(m, ' ');
 360                 seq_puts(m, name);
 361         }
 362         seq_putc(m, '\n');
 363 }
 364
 365 static int show_map(struct seq_file *m, void *v, int is_pid)
 366 {
 367         show_map_vma(m, v, is_pid);
 368         m_cache_vma(m, v);
 369         return 0;
 370 }
 371
 372 static int show_pid_map(struct seq_file *m, void *v)
 373 {
 374         return show_map(m, v, 1);
 375 }
 376
 377 static int show_tid_map(struct seq_file *m, void *v)
 378 {
 379         return show_map(m, v, 0);
 380 }
 381
 382 static const struct seq_operations proc_pid_maps_op = {
 383         .start  = m_start,
 384         .next   = m_next,
 385         .stop   = m_stop,
 386         .show   = show_pid_map
 387 };
 388
 389 static const struct seq_operations proc_tid_maps_op = {
 390         .start  = m_start,
 391         .next   = m_next,
 392         .stop   = m_stop,
 393         .show   = show_tid_map
 394 };
 395
 396 static int pid_maps_open(struct inode *inode, struct file *file)
 397 {
 398         return do_maps_open(inode, file, &proc_pid_maps_op);
 399 }
 400
 401 static int tid_maps_open(struct inode *inode, struct file *file)
 402 {
 403         return do_maps_open(inode, file, &proc_tid_maps_op);
 404 }
 405
 406 const struct file_operations proc_pid_maps_operations = {
 407         .open           = pid_maps_open,
 408         .read           = seq_read,
 409         .llseek         = seq_lseek,
 410         .release        = proc_map_release,
 411 };
 412
 413 const struct file_operations proc_tid_maps_operations = {
 414         .open           = tid_maps_open,
 415         .read           = seq_read,
 416         .llseek         = seq_lseek,
 417         .release        = proc_map_release,
 418 };
 419
 420 /*
 421  * Proportional Set Size(PSS): my share of RSS.
 422  *
 423  * PSS of a process is the count of pages it has in memory, where each
 424  * page is divided by the number of processes sharing it.  So if a
 425  * process has 1000 pages all to itself, and 1000 shared with one other
 426  * process, its PSS will be 1500.
 427  *
 428  * To keep (accumulated) division errors low, we adopt a 64bit
 429  * fixed-point pss counter to minimize division errors. So (pss >>
 430  * PSS_SHIFT) would be the real byte count.
 431  *
 432  * A shift of 12 before division means (assuming 4K page size):
 433  *      - 1M 3-user-pages add up to 8KB errors;
 434  *      - supports mapcount up to 2^24, or 16M;
 435  *      - supports PSS up to 2^52 bytes, or 4PB.
 436  */
 437 #define PSS_SHIFT 12
 438
 439 #ifdef CONFIG_PROC_PAGE_MONITOR
 440 struct mem_size_stats {
 441         unsigned long resident;
 442         unsigned long shared_clean;
 443         unsigned long shared_dirty;
 444         unsigned long private_clean;
 445         unsigned long private_dirty;
 446         unsigned long referenced;
 447         unsigned long anonymous;
 448         unsigned long anonymous_thp;
 449         unsigned long swap;
 450         unsigned long shared_hugetlb;
 451         unsigned long private_hugetlb;
 452         u64 pss;
 453         u64 swap_pss;
 454 };
 455
 456 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 457                 unsigned long size, bool young, bool dirty)
 458 {
 459         int mapcount;
 460
 461         if (PageAnon(page))
 462                 mss->anonymous += size;
 463
 464         mss->resident += size;
 465         /* Accumulate the size in pages that have been accessed. */
 466         if (young || page_is_young(page) || PageReferenced(page))
 467                 mss->referenced += size;
 468         mapcount = page_mapcount(page);
 469         if (mapcount >= 2) {
 470                 u64 pss_delta;
 471
 472                 if (dirty || PageDirty(page))
 473                         mss->shared_dirty += size;
 474                 else
 475                         mss->shared_clean += size;
 476                 pss_delta = (u64)size << PSS_SHIFT;
 477                 do_div(pss_delta, mapcount);
 478                 mss->pss += pss_delta;
 479         } else {
 480                 if (dirty || PageDirty(page))
 481                         mss->private_dirty += size;
 482                 else
 483                         mss->private_clean += size;
 484                 mss->pss += (u64)size << PSS_SHIFT;
 485         }
 486 }
 487
 488 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 489                 struct mm_walk *walk)
 490 {
 491         struct mem_size_stats *mss = walk->private;
 492         struct vm_area_struct *vma = walk->vma;
 493         struct page *page = NULL;
 494
 495         if (pte_present(*pte)) {
 496                 page = vm_normal_page(vma, addr, *pte);
 497         } else if (is_swap_pte(*pte)) {
 498                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 499
 500                 if (!non_swap_entry(swpent)) {
 501                         int mapcount;
 502
 503                         mss->swap += PAGE_SIZE;
 504                         mapcount = swp_swapcount(swpent);
 505                         if (mapcount >= 2) {
 506                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 507
 508                                 do_div(pss_delta, mapcount);
 509                                 mss->swap_pss += pss_delta;
 510                         } else {
 511                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 512                         }
 513                 } else if (is_migration_entry(swpent))
 514                         page = migration_entry_to_page(swpent);
 515         }
 516
 517         if (!page)
 518                 return;
 519         smaps_account(mss, page, PAGE_SIZE, pte_young(*pte), pte_dirty(*pte));
 520 }
 521
 522 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 523 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 524                 struct mm_walk *walk)
 525 {
 526         struct mem_size_stats *mss = walk->private;
 527         struct vm_area_struct *vma = walk->vma;
 528         struct page *page;
 529
 530         /* FOLL_DUMP will return -EFAULT on huge zero page */
 531         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 532         if (IS_ERR_OR_NULL(page))
 533                 return;
 534         mss->anonymous_thp += HPAGE_PMD_SIZE;
 535         smaps_account(mss, page, HPAGE_PMD_SIZE,
 536                         pmd_young(*pmd), pmd_dirty(*pmd));
 537 }
 538 #else
 539 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 540                 struct mm_walk *walk)
 541 {
 542 }
 543 #endif
 544
 545 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 546                            struct mm_walk *walk)
 547 {
 548         struct vm_area_struct *vma = walk->vma;
 549         pte_t *pte;
 550         spinlock_t *ptl;
 551
 552         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
 553                 smaps_pmd_entry(pmd, addr, walk);
 554                 spin_unlock(ptl);
 555                 return 0;
 556         }
 557
 558         if (pmd_trans_unstable(pmd))
 559                 return 0;
 560         /*
 561          * The mmap_sem held all the way back in m_start() is what
 562          * keeps khugepaged out of here and from collapsing things
 563          * in here.
 564          */
 565         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 566         for (; addr != end; pte++, addr += PAGE_SIZE)
 567                 smaps_pte_entry(pte, addr, walk);
 568         pte_unmap_unlock(pte - 1, ptl);
 569         cond_resched();
 570         return 0;
 571 }
 572
 573 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 574 {
 575         /*
 576          * Don't forget to update Documentation/ on changes.
 577          */
 578         static const char mnemonics[BITS_PER_LONG][2] = {
 579                 /*
 580                  * In case if we meet a flag we don't know about.
 581                  */
 582                 [0 ... (BITS_PER_LONG-1)] = "??",
 583
 584                 [ilog2(VM_READ)]        = "rd",
 585                 [ilog2(VM_WRITE)]       = "wr",
 586                 [ilog2(VM_EXEC)]        = "ex",
 587                 [ilog2(VM_SHARED)]      = "sh",
 588                 [ilog2(VM_MAYREAD)]     = "mr",
 589                 [ilog2(VM_MAYWRITE)]    = "mw",
 590                 [ilog2(VM_MAYEXEC)]     = "me",
 591                 [ilog2(VM_MAYSHARE)]    = "ms",
 592                 [ilog2(VM_GROWSDOWN)]   = "gd",
 593                 [ilog2(VM_PFNMAP)]      = "pf",
 594                 [ilog2(VM_DENYWRITE)]   = "dw",
 595 #ifdef CONFIG_X86_INTEL_MPX
 596                 [ilog2(VM_MPX)]         = "mp",
 597 #endif
 598                 [ilog2(VM_LOCKED)]      = "lo",
 599                 [ilog2(VM_IO)]          = "io",
 600                 [ilog2(VM_SEQ_READ)]    = "sr",
 601                 [ilog2(VM_RAND_READ)]   = "rr",
 602                 [ilog2(VM_DONTCOPY)]    = "dc",
 603                 [ilog2(VM_DONTEXPAND)]  = "de",
 604                 [ilog2(VM_ACCOUNT)]     = "ac",
 605                 [ilog2(VM_NORESERVE)]   = "nr",
 606                 [ilog2(VM_HUGETLB)]     = "ht",
 607                 [ilog2(VM_ARCH_1)]      = "ar",
 608                 [ilog2(VM_DONTDUMP)]    = "dd",
 609 #ifdef CONFIG_MEM_SOFT_DIRTY
 610                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 611 #endif
 612                 [ilog2(VM_MIXEDMAP)]    = "mm",
 613                 [ilog2(VM_HUGEPAGE)]    = "hg",
 614                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 615                 [ilog2(VM_MERGEABLE)]   = "mg",
 616                 [ilog2(VM_UFFD_MISSING)]= "um",
 617                 [ilog2(VM_UFFD_WP)]     = "uw",
 618         };
 619         size_t i;
 620
 621         seq_puts(m, "VmFlags: ");
 622         for (i = 0; i < BITS_PER_LONG; i++) {
 623                 if (vma->vm_flags & (1UL << i)) {
 624                         seq_printf(m, "%c%c ",
 625                                    mnemonics[i][0], mnemonics[i][1]);
 626                 }
 627         }
 628         seq_putc(m, '\n');
 629 }
 630
 631 #ifdef CONFIG_HUGETLB_PAGE
 632 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 633                                  unsigned long addr, unsigned long end,
 634                                  struct mm_walk *walk)
 635 {
 636         struct mem_size_stats *mss = walk->private;
 637         struct vm_area_struct *vma = walk->vma;
 638         struct page *page = NULL;
 639
 640         if (pte_present(*pte)) {
 641                 page = vm_normal_page(vma, addr, *pte);
 642         } else if (is_swap_pte(*pte)) {
 643                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 644
 645                 if (is_migration_entry(swpent))
 646                         page = migration_entry_to_page(swpent);
 647         }
 648         if (page) {
 649                 int mapcount = page_mapcount(page);
 650
 651                 if (mapcount >= 2)
 652                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 653                 else
 654                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 655         }
 656         return 0;
 657 }
 658 #endif /* HUGETLB_PAGE */
 659
 660 static int show_smap(struct seq_file *m, void *v, int is_pid)
 661 {
 662         struct vm_area_struct *vma = v;
 663         struct mem_size_stats mss;
 664         struct mm_walk smaps_walk = {
 665                 .pmd_entry = smaps_pte_range,
 666 #ifdef CONFIG_HUGETLB_PAGE
 667                 .hugetlb_entry = smaps_hugetlb_range,
 668 #endif
 669                 .mm = vma->vm_mm,
 670                 .private = &mss,
 671         };
 672
 673         memset(&mss, 0, sizeof mss);
 674         /* mmap_sem is held in m_start */
 675         walk_page_vma(vma, &smaps_walk);
 676
 677         show_map_vma(m, vma, is_pid);
 678
 679         seq_printf(m,
 680                    "Size:           %8lu kB\n"
 681                    "Rss:            %8lu kB\n"
 682                    "Pss:            %8lu kB\n"
 683                    "Shared_Clean:   %8lu kB\n"
 684                    "Shared_Dirty:   %8lu kB\n"
 685                    "Private_Clean:  %8lu kB\n"
 686                    "Private_Dirty:  %8lu kB\n"
 687                    "Referenced:     %8lu kB\n"
 688                    "Anonymous:      %8lu kB\n"
 689                    "AnonHugePages:  %8lu kB\n"
 690                    "Shared_Hugetlb: %8lu kB\n"
 691                    "Private_Hugetlb: %7lu kB\n"
 692                    "Swap:           %8lu kB\n"
 693                    "SwapPss:        %8lu kB\n"
 694                    "KernelPageSize: %8lu kB\n"
 695                    "MMUPageSize:    %8lu kB\n"
 696                    "Locked:         %8lu kB\n",
 697                    (vma->vm_end - vma->vm_start) >> 10,
 698                    mss.resident >> 10,
 699                    (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
 700                    mss.shared_clean  >> 10,
 701                    mss.shared_dirty  >> 10,
 702                    mss.private_clean >> 10,
 703                    mss.private_dirty >> 10,
 704                    mss.referenced >> 10,
 705                    mss.anonymous >> 10,
 706                    mss.anonymous_thp >> 10,
 707                    mss.shared_hugetlb >> 10,
 708                    mss.private_hugetlb >> 10,
 709                    mss.swap >> 10,
 710                    (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)),
 711                    vma_kernel_pagesize(vma) >> 10,
 712                    vma_mmu_pagesize(vma) >> 10,
 713                    (vma->vm_flags & VM_LOCKED) ?
 714                         (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
 715
 716         show_smap_vma_flags(m, vma);
 717         m_cache_vma(m, vma);
 718         return 0;
 719 }
 720
 721 static int show_pid_smap(struct seq_file *m, void *v)
 722 {
 723         return show_smap(m, v, 1);
 724 }
 725
 726 static int show_tid_smap(struct seq_file *m, void *v)
 727 {
 728         return show_smap(m, v, 0);
 729 }
 730
 731 static const struct seq_operations proc_pid_smaps_op = {
 732         .start  = m_start,
 733         .next   = m_next,
 734         .stop   = m_stop,
 735         .show   = show_pid_smap
 736 };
 737
 738 static const struct seq_operations proc_tid_smaps_op = {
 739         .start  = m_start,
 740         .next   = m_next,
 741         .stop   = m_stop,
 742         .show   = show_tid_smap
 743 };
 744
 745 static int pid_smaps_open(struct inode *inode, struct file *file)
 746 {
 747         return do_maps_open(inode, file, &proc_pid_smaps_op);
 748 }
 749
 750 static int tid_smaps_open(struct inode *inode, struct file *file)
 751 {
 752         return do_maps_open(inode, file, &proc_tid_smaps_op);
 753 }
 754
 755 const struct file_operations proc_pid_smaps_operations = {
 756         .open           = pid_smaps_open,
 757         .read           = seq_read,
 758         .llseek         = seq_lseek,
 759         .release        = proc_map_release,
 760 };
 761
 762 const struct file_operations proc_tid_smaps_operations = {
 763         .open           = tid_smaps_open,
 764         .read           = seq_read,
 765         .llseek         = seq_lseek,
 766         .release        = proc_map_release,
 767 };
 768
 769 enum clear_refs_types {
 770         CLEAR_REFS_ALL = 1,
 771         CLEAR_REFS_ANON,
 772         CLEAR_REFS_MAPPED,
 773         CLEAR_REFS_SOFT_DIRTY,
 774         CLEAR_REFS_MM_HIWATER_RSS,
 775         CLEAR_REFS_LAST,
 776 };
 777
 778 struct clear_refs_private {
 779         enum clear_refs_types type;
 780 };
 781
 782 #ifdef CONFIG_MEM_SOFT_DIRTY
 783 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 784                 unsigned long addr, pte_t *pte)
 785 {
 786         /*
 787          * The soft-dirty tracker uses #PF-s to catch writes
 788          * to pages, so write-protect the pte as well. See the
 789          * Documentation/vm/soft-dirty.txt for full description
 790          * of how soft-dirty works.
 791          */
 792         pte_t ptent = *pte;
 793
 794         if (pte_present(ptent)) {
 795                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 796                 ptent = pte_wrprotect(ptent);
 797                 ptent = pte_clear_soft_dirty(ptent);
 798                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 799         } else if (is_swap_pte(ptent)) {
 800                 ptent = pte_swp_clear_soft_dirty(ptent);
 801                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 802         }
 803 }
 804 #else
 805 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 806                 unsigned long addr, pte_t *pte)
 807 {
 808 }
 809 #endif
 810
 811 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 812 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 813                 unsigned long addr, pmd_t *pmdp)
 814 {
 815         pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
 816
 817         pmd = pmd_wrprotect(pmd);
 818         pmd = pmd_clear_soft_dirty(pmd);
 819
 820         if (vma->vm_flags & VM_SOFTDIRTY)
 821                 vma->vm_flags &= ~VM_SOFTDIRTY;
 822
 823         set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 824 }
 825 #else
 826 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 827                 unsigned long addr, pmd_t *pmdp)
 828 {
 829 }
 830 #endif
 831
 832 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 833                                 unsigned long end, struct mm_walk *walk)
 834 {
 835         struct clear_refs_private *cp = walk->private;
 836         struct vm_area_struct *vma = walk->vma;
 837         pte_t *pte, ptent;
 838         spinlock_t *ptl;
 839         struct page *page;
 840
 841         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
 842                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 843                         clear_soft_dirty_pmd(vma, addr, pmd);
 844                         goto out;
 845                 }
 846
 847                 page = pmd_page(*pmd);
 848
 849                 /* Clear accessed and referenced bits. */
 850                 pmdp_test_and_clear_young(vma, addr, pmd);
 851                 test_and_clear_page_young(page);
 852                 ClearPageReferenced(page);
 853 out:
 854                 spin_unlock(ptl);
 855                 return 0;
 856         }
 857
 858         if (pmd_trans_unstable(pmd))
 859                 return 0;
 860
 861         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 862         for (; addr != end; pte++, addr += PAGE_SIZE) {
 863                 ptent = *pte;
 864
 865                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
 866                         clear_soft_dirty(vma, addr, pte);
 867                         continue;
 868                 }
 869
 870                 if (!pte_present(ptent))
 871                         continue;
 872
 873                 page = vm_normal_page(vma, addr, ptent);
 874                 if (!page)
 875                         continue;
 876
 877                 /* Clear accessed and referenced bits. */
 878                 ptep_test_and_clear_young(vma, addr, pte);
 879                 test_and_clear_page_young(page);
 880                 ClearPageReferenced(page);
 881         }
 882         pte_unmap_unlock(pte - 1, ptl);
 883         cond_resched();
 884         return 0;
 885 }
 886
 887 static int clear_refs_test_walk(unsigned long start, unsigned long end,
 888                                 struct mm_walk *walk)
 889 {
 890         struct clear_refs_private *cp = walk->private;
 891         struct vm_area_struct *vma = walk->vma;
 892
 893         if (vma->vm_flags & VM_PFNMAP)
 894                 return 1;
 895
 896         /*
 897          * Writing 1 to /proc/pid/clear_refs affects all pages.
 898          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
 899          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
 900          * Writing 4 to /proc/pid/clear_refs affects all pages.
 901          */
 902         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
 903                 return 1;
 904         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
 905                 return 1;
 906         return 0;
 907 }
 908
 909 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
 910                                 size_t count, loff_t *ppos)
 911 {
 912         struct task_struct *task;
 913         char buffer[PROC_NUMBUF];
 914         struct mm_struct *mm;
 915         struct vm_area_struct *vma;
 916         enum clear_refs_types type;
 917         int itype;
 918         int rv;
 919
 920         memset(buffer, 0, sizeof(buffer));
 921         if (count > sizeof(buffer) - 1)
 922                 count = sizeof(buffer) - 1;
 923         if (copy_from_user(buffer, buf, count))
 924                 return -EFAULT;
 925         rv = kstrtoint(strstrip(buffer), 10, &itype);
 926         if (rv < 0)
 927                 return rv;
 928         type = (enum clear_refs_types)itype;
 929         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
 930                 return -EINVAL;
 931
 932         task = get_proc_task(file_inode(file));
 933         if (!task)
 934                 return -ESRCH;
 935         mm = get_task_mm(task);
 936         if (mm) {
 937                 struct clear_refs_private cp = {
 938                         .type = type,
 939                 };
 940                 struct mm_walk clear_refs_walk = {
 941                         .pmd_entry = clear_refs_pte_range,
 942                         .test_walk = clear_refs_test_walk,
 943                         .mm = mm,
 944                         .private = &cp,
 945                 };
 946
 947                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
 948                         /*
 949                          * Writing 5 to /proc/pid/clear_refs resets the peak
 950                          * resident set size to this mm's current rss value.
 951                          */
 952                         down_write(&mm->mmap_sem);
 953                         reset_mm_hiwater_rss(mm);
 954                         up_write(&mm->mmap_sem);
 955                         goto out_mm;
 956                 }
 957
 958                 down_read(&mm->mmap_sem);
 959                 if (type == CLEAR_REFS_SOFT_DIRTY) {
 960                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
 961                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
 962                                         continue;
 963                                 up_read(&mm->mmap_sem);
 964                                 down_write(&mm->mmap_sem);
 965                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
 966                                         vma->vm_flags &= ~VM_SOFTDIRTY;
 967                                         vma_set_page_prot(vma);
 968                                 }
 969                                 downgrade_write(&mm->mmap_sem);
 970                                 break;
 971                         }
 972                         mmu_notifier_invalidate_range_start(mm, 0, -1);
 973                 }
 974                 walk_page_range(0, ~0UL, &clear_refs_walk);
 975                 if (type == CLEAR_REFS_SOFT_DIRTY)
 976                         mmu_notifier_invalidate_range_end(mm, 0, -1);
 977                 flush_tlb_mm(mm);
 978                 up_read(&mm->mmap_sem);
 979 out_mm:
 980                 mmput(mm);
 981         }
 982         put_task_struct(task);
 983
 984         return count;
 985 }
 986
 987 const struct file_operations proc_clear_refs_operations = {
 988         .write          = clear_refs_write,
 989         .llseek         = noop_llseek,
 990 };
 991
 992 typedef struct {
 993         u64 pme;
 994 } pagemap_entry_t;
 995
 996 struct pagemapread {
 997         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
 998         pagemap_entry_t *buffer;
 999         bool show_pfn;
1000 };
1001
1002 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1003 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1004
1005 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1006 #define PM_PFRAME_BITS          55
1007 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1008 #define PM_SOFT_DIRTY           BIT_ULL(55)
1009 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1010 #define PM_FILE                 BIT_ULL(61)
1011 #define PM_SWAP                 BIT_ULL(62)
1012 #define PM_PRESENT              BIT_ULL(63)
1013
1014 #define PM_END_OF_BUFFER    1
1015
1016 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1017 {
1018         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1019 }
1020
1021 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1022                           struct pagemapread *pm)
1023 {
1024         pm->buffer[pm->pos++] = *pme;
1025         if (pm->pos >= pm->len)
1026                 return PM_END_OF_BUFFER;
1027         return 0;
1028 }
1029
1030 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1031                                 struct mm_walk *walk)
1032 {
1033         struct pagemapread *pm = walk->private;
1034         unsigned long addr = start;
1035         int err = 0;
1036
1037         while (addr < end) {
1038                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1039                 pagemap_entry_t pme = make_pme(0, 0);
1040                 /* End of address space hole, which we mark as non-present. */
1041                 unsigned long hole_end;
1042
1043                 if (vma)
1044                         hole_end = min(end, vma->vm_start);
1045                 else
1046                         hole_end = end;
1047
1048                 for (; addr < hole_end; addr += PAGE_SIZE) {
1049                         err = add_to_pagemap(addr, &pme, pm);
1050                         if (err)
1051                                 goto out;
1052                 }
1053
1054                 if (!vma)
1055                         break;
1056
1057                 /* Addresses in the VMA. */
1058                 if (vma->vm_flags & VM_SOFTDIRTY)
1059                         pme = make_pme(0, PM_SOFT_DIRTY);
1060                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1061                         err = add_to_pagemap(addr, &pme, pm);
1062                         if (err)
1063                                 goto out;
1064                 }
1065         }
1066 out:
1067         return err;
1068 }
1069
1070 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1071                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1072 {
1073         u64 frame = 0, flags = 0;
1074         struct page *page = NULL;
1075
1076         if (pte_present(pte)) {
1077                 if (pm->show_pfn)
1078                         frame = pte_pfn(pte);
1079                 flags |= PM_PRESENT;
1080                 page = vm_normal_page(vma, addr, pte);
1081                 if (pte_soft_dirty(pte))
1082                         flags |= PM_SOFT_DIRTY;
1083         } else if (is_swap_pte(pte)) {
1084                 swp_entry_t entry;
1085                 if (pte_swp_soft_dirty(pte))
1086                         flags |= PM_SOFT_DIRTY;
1087                 entry = pte_to_swp_entry(pte);
1088                 frame = swp_type(entry) |
1089                         (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1090                 flags |= PM_SWAP;
1091                 if (is_migration_entry(entry))
1092                         page = migration_entry_to_page(entry);
1093         }
1094
1095         if (page && !PageAnon(page))
1096                 flags |= PM_FILE;
1097         if (page && page_mapcount(page) == 1)
1098                 flags |= PM_MMAP_EXCLUSIVE;
1099         if (vma->vm_flags & VM_SOFTDIRTY)
1100                 flags |= PM_SOFT_DIRTY;
1101
1102         return make_pme(frame, flags);
1103 }
1104
1105 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1106                              struct mm_walk *walk)
1107 {
1108         struct vm_area_struct *vma = walk->vma;
1109         struct pagemapread *pm = walk->private;
1110         spinlock_t *ptl;
1111         pte_t *pte, *orig_pte;
1112         int err = 0;
1113
1114 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1115         if (pmd_trans_huge_lock(pmdp, vma, &ptl) == 1) {
1116                 u64 flags = 0, frame = 0;
1117                 pmd_t pmd = *pmdp;
1118
1119                 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd))
1120                         flags |= PM_SOFT_DIRTY;
1121
1122                 /*
1123                  * Currently pmd for thp is always present because thp
1124                  * can not be swapped-out, migrated, or HWPOISONed
1125                  * (split in such cases instead.)
1126                  * This if-check is just to prepare for future implementation.
1127                  */
1128                 if (pmd_present(pmd)) {
1129                         struct page *page = pmd_page(pmd);
1130
1131                         if (page_mapcount(page) == 1)
1132                                 flags |= PM_MMAP_EXCLUSIVE;
1133
1134                         flags |= PM_PRESENT;
1135                         if (pm->show_pfn)
1136                                 frame = pmd_pfn(pmd) +
1137                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1138                 }
1139
1140                 for (; addr != end; addr += PAGE_SIZE) {
1141                         pagemap_entry_t pme = make_pme(frame, flags);
1142
1143                         err = add_to_pagemap(addr, &pme, pm);
1144                         if (err)
1145                                 break;
1146                         if (pm->show_pfn && (flags & PM_PRESENT))
1147                                 frame++;
1148                 }
1149                 spin_unlock(ptl);
1150                 return err;
1151         }
1152
1153         if (pmd_trans_unstable(pmdp))
1154                 return 0;
1155 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1156
1157         /*
1158          * We can assume that @vma always points to a valid one and @end never
1159          * goes beyond vma->vm_end.
1160          */
1161         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1162         for (; addr < end; pte++, addr += PAGE_SIZE) {
1163                 pagemap_entry_t pme;
1164
1165                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1166                 err = add_to_pagemap(addr, &pme, pm);
1167                 if (err)
1168                         break;
1169         }
1170         pte_unmap_unlock(orig_pte, ptl);
1171
1172         cond_resched();
1173
1174         return err;
1175 }
1176
1177 #ifdef CONFIG_HUGETLB_PAGE
1178 /* This function walks within one hugetlb entry in the single call */
1179 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1180                                  unsigned long addr, unsigned long end,
1181                                  struct mm_walk *walk)
1182 {
1183         struct pagemapread *pm = walk->private;
1184         struct vm_area_struct *vma = walk->vma;
1185         u64 flags = 0, frame = 0;
1186         int err = 0;
1187         pte_t pte;
1188
1189         if (vma->vm_flags & VM_SOFTDIRTY)
1190                 flags |= PM_SOFT_DIRTY;
1191
1192         pte = huge_ptep_get(ptep);
1193         if (pte_present(pte)) {
1194                 struct page *page = pte_page(pte);
1195
1196                 if (!PageAnon(page))
1197                         flags |= PM_FILE;
1198
1199                 if (page_mapcount(page) == 1)
1200                         flags |= PM_MMAP_EXCLUSIVE;
1201
1202                 flags |= PM_PRESENT;
1203                 if (pm->show_pfn)
1204                         frame = pte_pfn(pte) +
1205                                 ((addr & ~hmask) >> PAGE_SHIFT);
1206         }
1207
1208         for (; addr != end; addr += PAGE_SIZE) {
1209                 pagemap_entry_t pme = make_pme(frame, flags);
1210
1211                 err = add_to_pagemap(addr, &pme, pm);
1212                 if (err)
1213                         return err;
1214                 if (pm->show_pfn && (flags & PM_PRESENT))
1215                         frame++;
1216         }
1217
1218         cond_resched();
1219
1220         return err;
1221 }
1222 #endif /* HUGETLB_PAGE */
1223
1224 /*
1225  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1226  *
1227  * For each page in the address space, this file contains one 64-bit entry
1228  * consisting of the following:
1229  *
1230  * Bits 0-54  page frame number (PFN) if present
1231  * Bits 0-4   swap type if swapped
1232  * Bits 5-54  swap offset if swapped
1233  * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1234  * Bit  56    page exclusively mapped
1235  * Bits 57-60 zero
1236  * Bit  61    page is file-page or shared-anon
1237  * Bit  62    page swapped
1238  * Bit  63    page present
1239  *
1240  * If the page is not present but in swap, then the PFN contains an
1241  * encoding of the swap file number and the page's offset into the
1242  * swap. Unmapped pages return a null PFN. This allows determining
1243  * precisely which pages are mapped (or in swap) and comparing mapped
1244  * pages between processes.
1245  *
1246  * Efficient users of this interface will use /proc/pid/maps to
1247  * determine which areas of memory are actually mapped and llseek to
1248  * skip over unmapped regions.
1249  */
1250 static ssize_t pagemap_read(struct file *file, char __user *buf,
1251                             size_t count, loff_t *ppos)
1252 {
1253         struct mm_struct *mm = file->private_data;
1254         struct pagemapread pm;
1255         struct mm_walk pagemap_walk = {};
1256         unsigned long src;
1257         unsigned long svpfn;
1258         unsigned long start_vaddr;
1259         unsigned long end_vaddr;
1260         int ret = 0, copied = 0;
1261
1262         if (!mm || !atomic_inc_not_zero(&mm->mm_users))
1263                 goto out;
1264
1265         ret = -EINVAL;
1266         /* file position must be aligned */
1267         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1268                 goto out_mm;
1269
1270         ret = 0;
1271         if (!count)
1272                 goto out_mm;
1273
1274         /* do not disclose physical addresses: attack vector */
1275         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1276
1277         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1278         pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1279         ret = -ENOMEM;
1280         if (!pm.buffer)
1281                 goto out_mm;
1282
1283         pagemap_walk.pmd_entry = pagemap_pmd_range;
1284         pagemap_walk.pte_hole = pagemap_pte_hole;
1285 #ifdef CONFIG_HUGETLB_PAGE
1286         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1287 #endif
1288         pagemap_walk.mm = mm;
1289         pagemap_walk.private = &pm;
1290
1291         src = *ppos;
1292         svpfn = src / PM_ENTRY_BYTES;
1293         start_vaddr = svpfn << PAGE_SHIFT;
1294         end_vaddr = mm->task_size;
1295
1296         /* watch out for wraparound */
1297         if (svpfn > mm->task_size >> PAGE_SHIFT)
1298                 start_vaddr = end_vaddr;
1299
1300         /*
1301          * The odds are that this will stop walking way
1302          * before end_vaddr, because the length of the
1303          * user buffer is tracked in "pm", and the walk
1304          * will stop when we hit the end of the buffer.
1305          */
1306         ret = 0;
1307         while (count && (start_vaddr < end_vaddr)) {
1308                 int len;
1309                 unsigned long end;
1310
1311                 pm.pos = 0;
1312                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1313                 /* overflow ? */
1314                 if (end < start_vaddr || end > end_vaddr)
1315                         end = end_vaddr;
1316                 down_read(&mm->mmap_sem);
1317                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1318                 up_read(&mm->mmap_sem);
1319                 start_vaddr = end;
1320
1321                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1322                 if (copy_to_user(buf, pm.buffer, len)) {
1323                         ret = -EFAULT;
1324                         goto out_free;
1325                 }
1326                 copied += len;
1327                 buf += len;
1328                 count -= len;
1329         }
1330         *ppos += copied;
1331         if (!ret || ret == PM_END_OF_BUFFER)
1332                 ret = copied;
1333
1334 out_free:
1335         kfree(pm.buffer);
1336 out_mm:
1337         mmput(mm);
1338 out:
1339         return ret;
1340 }
1341
1342 static int pagemap_open(struct inode *inode, struct file *file)
1343 {
1344         struct mm_struct *mm;
1345
1346         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1347         if (IS_ERR(mm))
1348                 return PTR_ERR(mm);
1349         file->private_data = mm;
1350         return 0;
1351 }
1352
1353 static int pagemap_release(struct inode *inode, struct file *file)
1354 {
1355         struct mm_struct *mm = file->private_data;
1356
1357         if (mm)
1358                 mmdrop(mm);
1359         return 0;
1360 }
1361
1362 const struct file_operations proc_pagemap_operations = {
1363         .llseek         = mem_lseek, /* borrow this */
1364         .read           = pagemap_read,
1365         .open           = pagemap_open,
1366         .release        = pagemap_release,
1367 };
1368 #endif /* CONFIG_PROC_PAGE_MONITOR */
1369
1370 #ifdef CONFIG_NUMA
1371
1372 struct numa_maps {
1373         unsigned long pages;
1374         unsigned long anon;
1375         unsigned long active;
1376         unsigned long writeback;
1377         unsigned long mapcount_max;
1378         unsigned long dirty;
1379         unsigned long swapcache;
1380         unsigned long node[MAX_NUMNODES];
1381 };
1382
1383 struct numa_maps_private {
1384         struct proc_maps_private proc_maps;
1385         struct numa_maps md;
1386 };
1387
1388 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1389                         unsigned long nr_pages)
1390 {
1391         int count = page_mapcount(page);
1392
1393         md->pages += nr_pages;
1394         if (pte_dirty || PageDirty(page))
1395                 md->dirty += nr_pages;
1396
1397         if (PageSwapCache(page))
1398                 md->swapcache += nr_pages;
1399
1400         if (PageActive(page) || PageUnevictable(page))
1401                 md->active += nr_pages;
1402
1403         if (PageWriteback(page))
1404                 md->writeback += nr_pages;
1405
1406         if (PageAnon(page))
1407                 md->anon += nr_pages;
1408
1409         if (count > md->mapcount_max)
1410                 md->mapcount_max = count;
1411
1412         md->node[page_to_nid(page)] += nr_pages;
1413 }
1414
1415 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1416                 unsigned long addr)
1417 {
1418         struct page *page;
1419         int nid;
1420
1421         if (!pte_present(pte))
1422                 return NULL;
1423
1424         page = vm_normal_page(vma, addr, pte);
1425         if (!page)
1426                 return NULL;
1427
1428         if (PageReserved(page))
1429                 return NULL;
1430
1431         nid = page_to_nid(page);
1432         if (!node_isset(nid, node_states[N_MEMORY]))
1433                 return NULL;
1434
1435         return page;
1436 }
1437
1438 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1439                 unsigned long end, struct mm_walk *walk)
1440 {
1441         struct numa_maps *md = walk->private;
1442         struct vm_area_struct *vma = walk->vma;
1443         spinlock_t *ptl;
1444         pte_t *orig_pte;
1445         pte_t *pte;
1446
1447         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1448                 pte_t huge_pte = *(pte_t *)pmd;
1449                 struct page *page;
1450
1451                 page = can_gather_numa_stats(huge_pte, vma, addr);
1452                 if (page)
1453                         gather_stats(page, md, pte_dirty(huge_pte),
1454                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1455                 spin_unlock(ptl);
1456                 return 0;
1457         }
1458
1459         if (pmd_trans_unstable(pmd))
1460                 return 0;
1461         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1462         do {
1463                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1464                 if (!page)
1465                         continue;
1466                 gather_stats(page, md, pte_dirty(*pte), 1);
1467
1468         } while (pte++, addr += PAGE_SIZE, addr != end);
1469         pte_unmap_unlock(orig_pte, ptl);
1470         return 0;
1471 }
1472 #ifdef CONFIG_HUGETLB_PAGE
1473 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1474                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1475 {
1476         pte_t huge_pte = huge_ptep_get(pte);
1477         struct numa_maps *md;
1478         struct page *page;
1479
1480         if (!pte_present(huge_pte))
1481                 return 0;
1482
1483         page = pte_page(huge_pte);
1484         if (!page)
1485                 return 0;
1486
1487         md = walk->private;
1488         gather_stats(page, md, pte_dirty(huge_pte), 1);
1489         return 0;
1490 }
1491
1492 #else
1493 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1494                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1495 {
1496         return 0;
1497 }
1498 #endif
1499
1500 /*
1501  * Display pages allocated per node and memory policy via /proc.
1502  */
1503 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1504 {
1505         struct numa_maps_private *numa_priv = m->private;
1506         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1507         struct vm_area_struct *vma = v;
1508         struct numa_maps *md = &numa_priv->md;
1509         struct file *file = vma->vm_file;
1510         struct mm_struct *mm = vma->vm_mm;
1511         struct mm_walk walk = {
1512                 .hugetlb_entry = gather_hugetlb_stats,
1513                 .pmd_entry = gather_pte_stats,
1514                 .private = md,
1515                 .mm = mm,
1516         };
1517         struct mempolicy *pol;
1518         char buffer[64];
1519         int nid;
1520
1521         if (!mm)
1522                 return 0;
1523
1524         /* Ensure we start with an empty set of numa_maps statistics. */
1525         memset(md, 0, sizeof(*md));
1526
1527         pol = __get_vma_policy(vma, vma->vm_start);
1528         if (pol) {
1529                 mpol_to_str(buffer, sizeof(buffer), pol);
1530                 mpol_cond_put(pol);
1531         } else {
1532                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1533         }
1534
1535         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1536
1537         if (file) {
1538                 seq_puts(m, " file=");
1539                 seq_file_path(m, file, "\n\t= ");
1540         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1541                 seq_puts(m, " heap");
1542         } else {
1543                 pid_t tid = pid_of_stack(proc_priv, vma, is_pid);
1544                 if (tid != 0) {
1545                         /*
1546                          * Thread stack in /proc/PID/task/TID/maps or
1547                          * the main process stack.
1548                          */
1549                         if (!is_pid || (vma->vm_start <= mm->start_stack &&
1550                             vma->vm_end >= mm->start_stack))
1551                                 seq_puts(m, " stack");
1552                         else
1553                                 seq_printf(m, " stack:%d", tid);
1554                 }
1555         }
1556
1557         if (is_vm_hugetlb_page(vma))
1558                 seq_puts(m, " huge");
1559
1560         /* mmap_sem is held by m_start */
1561         walk_page_vma(vma, &walk);
1562
1563         if (!md->pages)
1564                 goto out;
1565
1566         if (md->anon)
1567                 seq_printf(m, " anon=%lu", md->anon);
1568
1569         if (md->dirty)
1570                 seq_printf(m, " dirty=%lu", md->dirty);
1571
1572         if (md->pages != md->anon && md->pages != md->dirty)
1573                 seq_printf(m, " mapped=%lu", md->pages);
1574
1575         if (md->mapcount_max > 1)
1576                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1577
1578         if (md->swapcache)
1579                 seq_printf(m, " swapcache=%lu", md->swapcache);
1580
1581         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1582                 seq_printf(m, " active=%lu", md->active);
1583
1584         if (md->writeback)
1585                 seq_printf(m, " writeback=%lu", md->writeback);
1586
1587         for_each_node_state(nid, N_MEMORY)
1588                 if (md->node[nid])
1589                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1590
1591         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1592 out:
1593         seq_putc(m, '\n');
1594         m_cache_vma(m, vma);
1595         return 0;
1596 }
1597
1598 static int show_pid_numa_map(struct seq_file *m, void *v)
1599 {
1600         return show_numa_map(m, v, 1);
1601 }
1602
1603 static int show_tid_numa_map(struct seq_file *m, void *v)
1604 {
1605         return show_numa_map(m, v, 0);
1606 }
1607
1608 static const struct seq_operations proc_pid_numa_maps_op = {
1609         .start  = m_start,
1610         .next   = m_next,
1611         .stop   = m_stop,
1612         .show   = show_pid_numa_map,
1613 };
1614
1615 static const struct seq_operations proc_tid_numa_maps_op = {
1616         .start  = m_start,
1617         .next   = m_next,
1618         .stop   = m_stop,
1619         .show   = show_tid_numa_map,
1620 };
1621
1622 static int numa_maps_open(struct inode *inode, struct file *file,
1623                           const struct seq_operations *ops)
1624 {
1625         return proc_maps_open(inode, file, ops,
1626                                 sizeof(struct numa_maps_private));
1627 }
1628
1629 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1630 {
1631         return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1632 }
1633
1634 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1635 {
1636         return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1637 }
1638
1639 const struct file_operations proc_pid_numa_maps_operations = {
1640         .open           = pid_numa_maps_open,
1641         .read           = seq_read,
1642         .llseek         = seq_lseek,
1643         .release        = proc_map_release,
1644 };
1645
1646 const struct file_operations proc_tid_numa_maps_operations = {
1647         .open           = tid_numa_maps_open,
1648         .read           = seq_read,
1649         .llseek         = seq_lseek,
1650         .release        = proc_map_release,
1651 };
1652 #endif /* CONFIG_NUMA */